class Integrations::Openai::ProcessorService # 3.5 support 4,096 tokens # 1 token is approx 4 characters # 4,096 * 4 = 16,384 characters, sticking to 15,000 to be safe TOKEN_LIMIT = 15_000 API_URL = 'https://api.openai.com/v1/chat/completions'.freeze GPT_MODEL = 'gpt-3.5-turbo'.freeze ALLOWED_EVENT_NAMES = %w[rephrase summarize reply_suggestion].freeze pattr_initialize [:hook!, :event!] def perform event_name = event['name'] return nil unless valid_event_name?(event_name) send("#{event_name}_message") end private def valid_event_name?(event_name) ALLOWED_EVENT_NAMES.include?(event_name) end def rephrase_body { model: GPT_MODEL, messages: [ { role: 'system', content: "You are a helpful support agent. Please rephrase the following response to a more #{event['data']['tone']} tone. " \ "Reply in the user's language." }, { role: 'user', content: event['data']['content'] } ] }.to_json end def conversation_messages(in_array_format: false) conversation = find_conversation messages = init_messages_body(in_array_format) add_messages_until_token_limit(conversation, messages, in_array_format) end def find_conversation hook.account.conversations.find_by(display_id: event['data']['conversation_display_id']) end def add_messages_until_token_limit(conversation, messages, in_array_format) character_count = 0 conversation.messages.chat.reorder('id desc').each do |message| character_count, message_added = add_message_if_within_limit(character_count, message, messages, in_array_format) break unless message_added end messages end def add_message_if_within_limit(character_count, message, messages, in_array_format) if valid_message?(message, character_count) add_message_to_list(message, messages, in_array_format) character_count += message.content.length [character_count, true] else [character_count, false] end end def valid_message?(message, character_count) message.content.present? && character_count + message.content.length <= TOKEN_LIMIT end def add_message_to_list(message, messages, in_array_format) formatted_message = format_message(message, in_array_format) messages.prepend(formatted_message) end def init_messages_body(in_array_format) in_array_format ? [] : '' end def format_message(message, in_array_format) in_array_format ? format_message_in_array(message) : format_message_in_string(message) end def format_message_in_array(message) { role: (message.incoming? ? 'user' : 'assistant'), content: message.content } end def format_message_in_string(message) sender_type = message.incoming? ? 'Customer' : 'Agent' "#{sender_type} #{message.sender&.name} : #{message.content}\n" end def summarize_body { model: GPT_MODEL, messages: [ { role: 'system', content: 'Please summarize the key points from the following conversation between support agents and ' \ 'customer as bullet points for the next support agent looking into the conversation. Reply in the user\'s language.' }, { role: 'user', content: conversation_messages } ] }.to_json end def reply_suggestion_body { model: GPT_MODEL, messages: [ { role: 'system', content: 'Please suggest a reply to the following conversation between support agents and customer. Reply in the user\'s language.' } ].concat(conversation_messages(in_array_format: true)) }.to_json end def reply_suggestion_message make_api_call(reply_suggestion_body) end def summarize_message make_api_call(summarize_body) end def rephrase_message make_api_call(rephrase_body) end def make_api_call(body) headers = { 'Content-Type' => 'application/json', 'Authorization' => "Bearer #{hook.settings['api_key']}" } response = HTTParty.post(API_URL, headers: headers, body: body) JSON.parse(response.body)['choices'].first['message']['content'] end end